Thera Bank Personal Loan Campaign¶
Data Description:¶
The dataset contains data on 5000 customers. The data include customer demographic information (age, income, etc.), the customer's relationship with the bank (mortgage, securities account, etc.), and the customer response to the last personal loan campaign (Personal Loan). Among these 5000 customers, only 480 (= 9.6%) accepted the personal loan that was offered to them in the earlier campaign. Domain: Banking
Context:¶
This case is about a bank (Thera Bank) whose management wants to explore ways of converting its liability customers to personal loan customers (while retaining them as depositors). A campaign that the bank ran last year for liability customers showed a healthy conversion rate of over 9% success. This has encouraged the retail marketing department to devise campaigns with better target marketing to increase the success ratio with a minimal budget. Attribute Information:
- ID: Customer ID
- Age: Customer's age in completed years
- Experience: #years of professional experience
- Income: Annual income of the customer (000)
- ZIP Code: Home Address ZIP code.
- Family: Family size of the customer
- CCAvg: Avg. spending on credit cards per month (000)
- Education: Education Level. 1: Undergrad; 2: Graduate; 3: Advanced/Professional
- Mortgage: Value of house mortgage if any. ($000)
- Personal Loan: Did this customer accept the personal loan offered in the last campaign?
- Securities Account: Does the customer have a securities account with the bank?
- CD Account: Does the customer have a certificate of deposit (CD) account with the bank?
- Online: Does the customer use internet banking facilities?
- Credit card: Does the customer use a credit card issued by the bank?
Learning Outcomes:¶
- Exploratory Data Analysis
- Preparing the data to train a model
- Training and making predictions using a classification model
- Model evaluation ### Objective: The classification goal is to predict the likelihood of a liability customer buying personal loans.
Steps and tasks:¶
1. Import the datasets and libraries, check datatype, statistical summary, shape, null values or incorrect imputation. (5 marks)¶
In [1]:
import warnings
warnings.filterwarnings('ignore')
#Load Libraries
#!pip install pandas_profiling
import pandas as pd
import pandas_profiling
import numpy as np
import seaborn as sns
from sklearn import preprocessing
from sklearn.model_selection import train_test_split
from sklearn.linear_model import LogisticRegression
from sklearn.feature_selection import RFE
from sklearn import metrics
from sklearn.metrics import classification_report, confusion_matrix,recall_score, precision_score, f1_score, roc_auc_score,accuracy_score
import matplotlib.pyplot as plt
%matplotlib inline
# Remove scientific notations and display numbers with 2 decimal points instead
pd.options.display.float_format = '{:,.2f}'.format
In [2]:
#Load data
thera_df = pd.read_csv('Bank_Personal_Loan_Modelling.csv')
In [3]:
thera_df.head()
Out[3]:
In [4]:
# clean column names
thera_df.columns = ["ID","Age","Experience","Income","ZIPCode","Family","CCAvg","Education","Mortgage","PersonalLoan","SecuritiesAccount","CDAccount","Online","CreditCard"]
In [5]:
thera_df.shape
Out[5]:
In [6]:
thera_df.dtypes
Out[6]:
In [7]:
thera_df.isna().sum()
Out[7]:
In [8]:
thera_df.describe().transpose()
Out[8]:
2. EDA: Study the data distribution in each attribute and target variable, share your findings (20 marks)¶
- Number of unique in each column?
- Number of people with zero mortgage?
- Number of people with zero credit card spending per month?
- Value counts of all categorical columns.
- Univariate and Bivariate
- Get data model ready
In [9]:
#Number of unique in each column
thera_df.nunique()
Out[9]:
In [10]:
#Number of people with zero mortgage
len(thera_df[thera_df.Mortgage==0])
Out[10]:
In [11]:
#Number of people with zero credit card spending per month?
len(thera_df[thera_df.CCAvg==0])
Out[11]:
In [12]:
#Value counts of all categorical columns.
#family and Education are categorical
thera_df.Education.value_counts()
Out[12]:
In [13]:
thera_df.Family.value_counts()
Out[13]:
In [ ]:
In [ ]:
In [14]:
#Univariate and Bivariate
In [15]:
thera_df.profile_report()
Out[15]:
In [16]:
# negative values in Experince
thera_df.loc[(thera_df['Experience'] < 0), 'Experience'] = np.nan
# replace by median
thera_df['Experience'].fillna(thera_df['Experience'].median(),inplace=True)
In [17]:
len(thera_df[thera_df['Experience'] < 0])
Out[17]:
In [18]:
thera_df.columns
Out[18]:
In [19]:
cont_cols=['Age','Experience','Income','CCAvg','Mortgage']
thera_df[cont_cols]
indx=1
plt.figure(figsize= (40.5,40.5))
for col in cont_cols:
plt.subplot(5,2,indx)
plt.hist(thera_df[col])
plt.xlabel(col)
indx=indx+1
plt.subplot(5,2,indx)
sns.boxplot(x= thera_df[col])
indx=indx+1
Observation¶
Age seems to be quite normally distributed with majority of 35 and 55 years of age
Experience also normally distibuted with 11 to 30 q1 and q3
Income, CC Average and Mortgage are highly skewed
Significant outliers in mortagae and CCAvg
In [20]:
cat_cols=['Family',
'Education', 'SecuritiesAccount',
'CDAccount', 'Online', 'CreditCard']
indx=1
plt.figure(figsize= (30,50))
for col in cat_cols:
plt.subplot(6,2,indx)
thera_df[col].value_counts().plot(kind="bar", align='center',edgecolor = 'black')
plt.xlabel(col)
indx=indx+1
Observation:¶
1. Family and education are quite evenly distributed
2. CD , CC and Securities have vast difference
3. Online shopping also have significant difference in populationTop 5 Zip Codes¶
In [21]:
top5zipcode = thera_df[thera_df['PersonalLoan']==1]['ZIPCode'].value_counts().head(5)
top5zipcode
Out[21]:
In [22]:
thera_df.columns
Out[22]:
In [23]:
sns.countplot(x="CDAccount", data=thera_df,hue="PersonalLoan")
sns.countplot(x="CreditCard", data=thera_df,hue="PersonalLoan")
Out[23]:
In [24]:
sns.distplot( thera_df[thera_df['PersonalLoan'] == 0]['CCAvg'], color = 'r')
sns.distplot( thera_df[thera_df['PersonalLoan'] == 1]['CCAvg'], color = 'g')
Out[24]:
Observation¶
Customers with higher credit card average have taken more personal loan have .
In [25]:
sns.distplot( thera_df[thera_df.PersonalLoan == 0]['Income'], color = 'r')
sns.distplot( thera_df[thera_df.PersonalLoan == 1]['Income'], color = 'g')
Out[25]:
Observation¶
Customers with higher income have taken more personal loan have .
In [26]:
pairplt = sns.pairplot(thera_df[['Age','Experience','Income','ZIPCode','Family','CCAvg' ,'Education' , 'Mortgage','PersonalLoan','SecuritiesAccount','CDAccount','Online','CreditCard']] )
Observation¶
- Most of the customer's age is in the range of 30 to 60 yrs
- Most experience falls in the range of 5 to 35 years
- Most earn an income between 10K to 100K.
- Age and Experience has linear relation
In [27]:
# Corelation comparison
In [28]:
plt.figure(figsize=(25, 25))
ax = sns.heatmap(thera_df.corr(), fmt='.2f', annot=True, linecolor='white',linewidths=0.01,square=True)
plt.title('Correlation')
plt.show()
In [ ]:
Observation¶
The data set got 0 missing or null values
- Numeric variables: ‘Age’, ‘CC_Avg’, ‘ID’, ‘Income’, ‘Mortgage’, ‘Zip_Code’, ‘Experience’ Categorical variables: ‘Education’, ‘Family’
- Boolean variables: ‘CD_Account’, ‘Credit_Card’, ‘Online’, ‘Personal_Loan’, ‘Securities Account’
Column 'Experience' has negative values
Correlation:
- Experience is highly correlated with Age (0.99)
- Personal Loan is significantly correlated with Income, Credit cards sepending and certificate of deposit (CD) account.
- Income and CCAvg also corelated
We can drop ‘ID’, ‘Experience’ columns for further analysis as ‘ID’ is a series & ‘Experience’ and ‘Age’ are highly correlated.
In [29]:
#Get data model ready
df = thera_df.copy()
thera_df.head()
df.head(2)
#Drop 'ID', 'ZIP Code' & 'Experience' and move 'Personal Loan' to last column
y= df['PersonalLoan']
df.drop(['PersonalLoan'], axis = 1,inplace = True)
df.drop(['Experience'], axis = 1,inplace = True)
df['PersonalLoan'] = y
df.drop(['ID'], axis = 1,inplace = True)
df.head(2)
Out[29]:
In [30]:
#Look at the data distribution
df.describe().transpose()
Out[30]:
3. Split the data into training and test set in the ratio of 70:30 respectively (5 marks)¶
In [31]:
array = df
X= array.iloc[:,0:11]
y= array.iloc[:,11]
In [32]:
#Convert categorical vriables to dummy variables
X = pd.get_dummies(X, drop_first=True)
X.head(2)
Out[32]:
In [33]:
y.head(2)
Out[33]:
In [34]:
print('df', df.shape)
print('X', X.shape)
print("y",y.shape)
In [35]:
X.head()
Out[35]:
In [36]:
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.30, random_state=42)
In [37]:
print('X_train', X_train.shape)
print("X_test",X_test.shape)
print('y_train', y_train.shape)
print("y_test",y_test.shape)
4. Use Logistic Regression model to predict the number of a customers buying personal loans. Print all the metrics related for evaluating the model performance (15 marks)¶
In [38]:
## function to get confusion matrix in a proper format
def draw_cm( actual, predicted ):
cm = confusion_matrix( actual, predicted)
sns.heatmap(cm, annot=True, fmt='.2f', xticklabels = [0,1] , yticklabels = [0,1] )
plt.ylabel('Observed')
plt.xlabel('Predicted')
plt.show()
In [39]:
logreg = LogisticRegression()
logreg.fit(X_train, y_train)
y_pred = logreg.predict(X_test)
print('Accuracy of logistic regression classifier on train set: {:.2f}'.format(logreg.score(X_train, y_train)))
print('Accuracy of logistic regression classifier on test set: {:.2f}'.format(logreg.score(X_test, y_test)))
print("Trainig accuracy",logreg.score(X_train,y_train))
print()
print("Testing accuracy",logreg.score(X_test, y_test))
print()
print('Confusion Matrix')
print(draw_cm(y_test,y_pred))
print()
print("Recall:",recall_score(y_test,y_pred))
print()
print("Precision:",precision_score(y_test,y_pred))
print()
print("F1 Score:",f1_score(y_test,y_pred))
print()
print("Roc Auc Score:",roc_auc_score(y_test,y_pred))
In [40]:
df_confusion_matrix = confusion_matrix(y_test, y_pred)
print(df_confusion_matrix)
In [41]:
print(classification_report(y_test, y_pred))
In [ ]:
In [ ]:
In [42]:
#AUC ROC curve
from sklearn.metrics import roc_auc_score
from sklearn.metrics import roc_curve
logit_roc_auc = roc_auc_score(y_test, logreg.predict(X_test))
fpr, tpr, thresholds = roc_curve(y_test, logreg.predict_proba(X_test)[:,1])
plt.figure()
plt.plot(fpr, tpr, label='Logistic Regression (area = %0.2f)' % logit_roc_auc)
plt.plot([0, 1], [0, 1],'r--')
plt.xlim([0.0, 1.0])
plt.ylim([0.0, 1.05])
plt.xlabel('False Positive Rate')
plt.ylabel('True Positive Rate')
plt.title('Receiver operating characteristic')
plt.legend(loc="lower right")
plt.savefig('Log_ROC')
plt.show()
In [43]:
# Checking Parameters of logistic regression
logreg.get_params()
Out[43]:
5. Give your reasoning on how can the model perform better? (10 marks)¶
Hint: Check parameter
Try dropping ZIP CODE AS THEY ARE SERIES DATA
- MORE TRAINING DATA
- OUTLIER TREATMENT
- MISSING VLAUES
- Try MULTIPLE ALGORITHMS
- Try ALGORITHM TUNING
Checking Parameters of Logistic Regression¶
Here recall is more important then accuracy. As the recall is low at 32 and false negatives are high at 107, we will try to optimize the Logistic regression parameters
In [44]:
# Running a loop to check different values of 'solver'
# all solver can be used with l2, only 'liblinear' and 'saga' works with both 'l1' and 'l2'
train_score=[]
test_score=[]
recall_score_list=[]
solver = ['newton-cg','lbfgs','liblinear','sag','saga']
for i in solver:
model = LogisticRegression(random_state=42,penalty='l2', C = 0.75,solver=i) # changing values of solver
model.fit(X_train, y_train)
y_predict = model.predict(X_test)
train_score.append(round(model.score(X_train, y_train),3))
test_score.append(round(model.score(X_test, y_test),3))
recall_score_list.append(round(recall_score(y_test,y_predict),3))
print(solver)
print()
print(train_score)
print()
print(test_score)
print()
print(recall_score_list)
In [45]:
train_score=[]
test_score=[]
recall_score_list=[]
solver = ['liblinear','saga'] # changing values of solver which works with 'l1'
for i in solver:
model = LogisticRegression(random_state=42,penalty='l1', C = 0.75,solver=i) #changed penalty to 'l1'
model.fit(X_train, y_train)
y_predict = model.predict(X_test)
train_score.append(round(model.score(X_train, y_train),3))
test_score.append(round(model.score(X_test, y_test),3))
recall_score_list.append(round(recall_score(y_test,y_predict),3))
print(solver)
print()
print(train_score)
print()
print(test_score)
print()
print(recall_score_list)
In [46]:
## Highest accuracy and recall of 'l1' with 'liblinear' are little better 'l2' with 'newton-cg'
# choosing liblinear with 'l1'
In [47]:
model = LogisticRegression(random_state=42,penalty='l1',solver='liblinear',class_weight='balanced') # changing class weight to balanced
model.fit(X_train, y_train)
y_predict = model.predict(X_test)
print("Trainig accuracy",model.score(X_train,y_train))
print()
print("Testing accuracy",model.score(X_test, y_test))
print()
print("Recall:",recall_score(y_test,y_predict))
In [48]:
# Running a loop to check different values of 'C'
train_score=[]
test_score=[]
recall_score_list=[]
C = [0.01,0.1,0.25,0.5,0.75,1]
for i in C:
model = LogisticRegression(random_state=42,penalty='l1', solver='liblinear',class_weight='balanced', C=i) # changing values of C
model.fit(X_train, y_train)
y_predict = model.predict(X_test)
train_score.append(round(model.score(X_train,y_train),3)) # appending training accuracy in a blank list for every run of the loop
test_score.append(round(model.score(X_test, y_test),3)) # appending testing accuracy in a blank list for every run of the loop
recall_score_list.append(round(recall_score(y_test,y_predict),3))
print(C)
print()
print(train_score)
print()
print(test_score)
print()
print(recall_score_list)
In [49]:
#Therefore final model is
model = LogisticRegression(random_state=42,penalty='l1', solver='liblinear', class_weight='balanced',C=0.25)
model.fit(X_train, y_train)
y_predict = model.predict(X_test)
print("Trainig accuracy",model.score(X_train,y_train))
print()
print("Testing accuracy",model.score(X_test, y_test))
print()
print('Confusion Matrix')
print(draw_cm(y_test,y_predict))
print()
print("Recall:",recall_score(y_test,y_predict))
print()
print("Precision:",precision_score(y_test,y_predict))
print()
print("F1 Score:",f1_score(y_test,y_predict))
print()
print("Roc Auc Score:",roc_auc_score(y_test,y_predict))
In [50]:
cm = confusion_matrix( y_test,y_predict)
print(cm)
Result of logistric regression's parameters tuning¶
- Recall improved to 91 from 32
- Roc score to 90 from 64 %
6. Give Business understanding of your model? (5 marks)¶
The campaign's goal is to have more people to accept personal loan i.e. less number of False Negative, so it don't lose customers who want to take personal loan.
Types of error:
- Type I error — False postives : Predicted that Customer has taken personal loan but actually customer has not.
- Type II error — False negatives : Predicted that Customer has not taken personal loan but actually customer has.
In this use case, we need to reduce Type II error to not lose the people who would take loans. Hence, Recall is MORE important then precision for our usecase.
In input data, the number of buyer’s percentage to the non-buyer percentage is very less. Hence, the accuracy is not an important performance criteria.
So, the right model for this usecase has to have minimize Type II error and have high recall (low false negative)
Final model has recall score of 91 % up from initial 32% and the false negative are down from 107 to 13.
After achieving the desired accuracy we can deploy the model for practical use. As in the bank now can predict who is interested in personal loans. They can use the model for upcoming campaigns and customers.
In [ ]: